• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

why don't we hate rambus?

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.


<<

<< Sorry, but I'm a firm NON-believer in consumer activism. It's silly, and accomplishes nothing, but denying yourself the full market to choose from. >>


That line just doesn't work. Up in Toronto, kids started boycotting McDonald's because they used Styrofoam packaging (a long time ago). Guess what? They switched to paper.

We still have the god damned Big Mac, eh? >>



Public relations... I guarantee that McDonald's profits did not suffer.

Here's a kicker: The Black boycott of Nike shoes back in the mid ninties. That one was funny. Sales JUMPED right after it was declared.

Please name one company that folded by not reacting to a boycott. Even IF you could find one, I'll show you dozens upon dozens that did not.
 
The issue is not that they didn't have a right to claim royalties if they wanted to. They made the tech, they had every right to say "You can use this, but you'll need to pay us a small fee for it."

Instead of doing that, they went to the JEDEC and said "You can use this, make it part of your standard, which we know is an open, royalty-free standard." Then years later they came out and said "Hey, you guys are using our patented technology, you owe us lots of money," after the standard was entrenched and there was no way that memory makers could say "well, we just won't use your tech anymore". This is called fraud, deliberate fraud.
 


<< The issue is not that they didn't have a right to claim royalties if they wanted to. They made the tech, they had every right to say "You can use this, but you'll need to pay us a small fee for it."

Instead of doing that, they went to the JEDEC and said "You can use this, make it part of your standard, which we know is an open, royalty-free standard." Then years later they came out and said "Hey, you guys are using our patented technology, you owe us lots of money," after the standard was entrenched and there was no way that memory makers could say "well, we just won't use your tech anymore". This is called fraud, deliberate fraud. >>



Doesn't sound like fraud to me. Bad timing, questionable ethics? Maybe. But, as long as they did not give up the rights to their patents, they own them.

Where in the patent law does it say you have to charge royalties from the beginning? Where dsoes it say you can't start charging royalties after an item becomes popular?

If there is a clause for this, I'll capitulate. 🙂
 
Bottom line: if you ask for user perference on these boards, 90+ % would take an Athlon XP with DDR over a P4 with Rambus (RDRAM). We may not hate Rambus with the passion we once had, but there is no love for the company either.

They lost. We won. We're over it.
 


<< "It constantly amazes me how some people could regard one company as "evil" and another as "good". As if they're not all run by the same kind of greedy board of directors. You think if AMD had 75% of the CPU market that they'll still be selling their processors for as cheap as they could? Remember back when the first 1GHz Athlon was released and how much it cost? RDRAM is a patented technology run by one company who happened to have gotten in Intel's favor and felt they could afford to try to expand through the courts. Big deal, it's business, not exactly the most moral field of industry in the world. Thinking that one company is somehow "moral" and the other is not is just diluded and smells of zealoutry."

You would love _The Tragedy of Great Power Politics_ by John Mearshimer. He argues that all states are exactly the same: they want security and they'll wage offensive wars to get them.

Here I paraphrase Winston Churchill: [AMD] is the worst possible [company], except for all the others. Sure they're bad, but to say that all companies are going to try to corner their market by half-legal means--or to say that any tech company would try to corner the dram market by claiming it had patents to JEDEC open standards strikes me as absolute BS. There are some people who have a better moral standard (as measured by the degree it helps the human condition) than others, and there are some companies that have better moral standards than others. The real question is which is responsible for more wrong: government or industries.

This has diverged somewhat from the original line of inquiry, so . . .

Competition is good. Rambus tried to squelch competition.
AMD, by comparison, has not tried to squelch competition, and whenever they make a profit they have pretty good profit sharing amongst its employees.
Intel isn't bad, per say, but they've engaged in some shady business practices, e.g. buying into memory companies to help the Rambus cause along.
Microsoft. Is bad.

Maybe this is the natural result of being on top--that a company tries to one-up its up and coming competition--but I don't think this is necessarily the case. cf IBM, Coke, Motorola, nvidia.

Nvidia offers a particularly compelling example. With the one exception of the (awful) memo about how to turn people away from the kyro II, they've been more or less fine upstanding citizens. You can use a Radeon on an nforce, for instance. They've yielded contracts to Dell without doing stuff like dumping. >>



nVidia is not on top. Their company capital is no where near enough to dwarf ATI, Coke is run by a bigger corporation and does have a lot of competition, Motorola is by no where near top of the hill as far as in the market. That asside, businesses will do what they can to stay on top. Rambus did some things that aren't "approved" by a lot of people (and exaggerated beyond belief by most) but it's a business practice. Avoiding a company's product simply because you don't agree with their business practices is utter non-sense and believing that somehow one company is more "moral" is certainly rediculous. One company simply decided that they do not wish to go that route in order to gain an advantage, others did, in the end, it's all business. If AMD could gain ground by suing Intel for some patent infringement (and I'm not sure, but I do think this happened) you better believe they'd take it. I don't know about you but I have no doubt that if AMD could gain total market control, that they wouldn't. Morality is not all about actions, it's about intent. As for reputation, Rambus saw an opportunity and it decided to take it. Other companys either didn't see an opportunity to wipe one of its competitors or decided it was too risky. If it was a sure-fire thing to take up a direct competitor do you honestly think any company wouldn't?
 


<< Doesn't sound like fraud to me. Bad timing, questionable ethics? Maybe. But, as long as they did not give up the rights to their patents, they own them. >>



Again, if they had done it properly, I'd be fine with it. But they did NOT submit the technology to be included in the JEDEC standard with any mention of it being patented. They submitted it as part of an OPEN NON-ROYALTY standard. They intentionally hid the fact that they had any patents on it at all (if you were on the committee and somebody said "here, use this for free" would you go digging through patents to see if they had any on it, or assume they were acting in good faith?). What they did was both fraud by representing it as a non-patented (or at least non-royalty) technology that could be used in the open standard, when they intended to attempt to get royalties on it (internal information did show they knew what they were doing), and it was acting in extremely bad faith. I can't imagine anybody accepting Rambus as part of a technology committee again (I wouldn't, but somebody probably will).

Both the courts and nearly all other technology companies agree that Rambus committed fraud here. While you certainly have a right to disagree, I am completely unable to see how you can think that what they did wasn't a deliberate attempt to defraud.
 


<< The issue is not that they didn't have a right to claim royalties if they wanted to. They made the tech, they had every right to say "You can use this, but you'll need to pay us a small fee for it."

Instead of doing that, they went to the JEDEC and said "You can use this, make it part of your standard, which we know is an open, royalty-free standard." Then years later they came out and said "Hey, you guys are using our patented technology, you owe us lots of money," after the standard was entrenched and there was no way that memory makers could say "well, we just won't use your tech anymore". This is called fraud, deliberate fraud. >>



First, that is not fraud. Second, being a person who read the court transcipts and followed the trial day by day.. opinions vary WIDELY in this. A theory of pro-Rambusians (disclosure: that includes me) was that Rambus got invited by JEDEC. They showed the committee their technology under Non-disclosure Agreements in which JEDEC cheated Rambus of their technology. I could go on and on forever about this, but I will just put it in a nutshell. Companies used Rambus's knowledge to spin off/create SDRAM and DDR RAM. They nit-picked Rambus's IP to create something so they would not have to pay Rambus rolyalties on RDRAM. They were extemely intelligent and good in doing so too. When Intel gave money to Samsung and Micron to invest in RDRAM how-to, Micron decided to use the money for other things. They were in the TEAM DDR campaign. Rambus never wanted to sue, but they got gipped and they are a public company who has owners, they had to. It was their right. Documents uncovered during the Infineon Trial had uncovered a memo that stated, "Someday all memory will be created this way, hopefully without royalties." BTW, the judge threw out the DDR fraud later on, and the other court cases someone had state they are losing. I don't know where they are getting that, considering most of the rulings and the independant experts sides Rambus in the dealings. I know this is a scattered timeline, but I do not have the time to uncover dates and events to explain step by step on what happened. Just what I know from observation and discussions. JMHO.
 


<<

<< The issue is not that they didn't have a right to claim royalties if they wanted to. They made the tech, they had every right to say "You can use this, but you'll need to pay us a small fee for it."

Instead of doing that, they went to the JEDEC and said "You can use this, make it part of your standard, which we know is an open, royalty-free standard." Then years later they came out and said "Hey, you guys are using our patented technology, you owe us lots of money," after the standard was entrenched and there was no way that memory makers could say "well, we just won't use your tech anymore". This is called fraud, deliberate fraud. >>



Doesn't sound like fraud to me. Bad timing, questionable ethics? Maybe. But, as long as they did not give up the rights to their patents, they own them.

Where in the patent law does it say you have to charge royalties from the beginning? Where dsoes it say you can't start charging royalties after an item becomes popular?

If there is a clause for this, I'll capitulate. 🙂 >>




You are right on the patent law. As soon as Rambus found out they were changing their IP so slightly so they wouldn't have to pay royalties. Rambus amended their original patent. This is completely legal too. Also, Rambus had left JEDEC before ameding their patent.
 
this thread is quite long so i'm not sure that anyone will read this. but it is important which we choose, because what we choose to use today has a lot to do with what we will get to buy in the future. if we had all left SDRAM and now DDR back then, Most of research dollars would have gone to rambus and DDR / QDR would have died. because we chose to stay with SDRAM / DDR, Research dollars have continued to flow in that direction and we are likely to see the end of RDRAM and the beginning of QDRAM. obviously this is an oversimplification of a VERY complex process but i'm sure most of us are intelligent enough to see that there is truth even in the simplicity. 🙂
 
""Instead of doing that, they went to the JEDEC and said "You can use this, make it part of your standard, which we know is an open, royalty-free standard." Then years later they came out and said "Hey, you guys are using our patented technology, you owe us lots of money," after the standard was entrenched and there was no way that memory makers could say "well, we just won't use your tech anymore". This is called fraud, deliberate fraud. ""

I agree with you, but in addition, i do not think alot of people seem to understand that when you bring furth a standard in JEDEC, you legally agree to make it public domain. You do not bring IP you wish to protect to a JEDEC meeting. Even if Rambus did invent all the memory IP we use today, the point is very mute as they agreed to make it public domain by going to a industry standards entitity like JEDEC. Either Rmabus was very stupid ( unlikely ) or they where where actually making a fairly good assumption that the legal system would get confused by the issue ( which is very likely with our legal system when it comes to the application of IP ). unfortunatly for Rambus, it under estimated how much the industry disliked it. Quite simply Rambus failed to do some research into attempting to push a propritary standard into the comsumer PC market. Unfortuntly Rambus failed to realize that to introduce a propritary standard it must significantly satisfy the need for one of the follwing, Performance, price, or diar need. of which it's Rambus accomplished none of them.

-It's performance is bascially equal to DDR, If you took every single benchmark on the subject possible you will almost certainly find that the diffarnce in perforamce on average is not even worth speaking about

-it's price is not cheaper than DDR and most of the time has been significantly higher. Sure today it's priced within reason, but given point 1, there is not compelling reason for buy ram at the same price for something that performs the same. If it was cheaper than DDR ram, we might have something to work on, but as of now, there is no evidence this will ever happen other than when the day comes that shops are trying to dump their enventory of rambus that nobody wants.

-And there is no need for it in the futre despite claims by some that it will become more disirable when proc speeds increase. Poping in an imaginary 10Ghz proc will not magically make Rambus suddenly shine, in fact it will just make it look worse as the latency penalties will cripple the the enivitable longer pipeline a 10Ghz proc will need. the longer the pipeline gets, the more it can be crippled by high latencies weather it be Rambus or DDR. Which BTW is touching on the major weakness of the P4 in general, the more complexe the task you trough at it, the slower it gets since the whole thing is a high latency model of computing. Sure, give it a simple task like a media file encoding, it will do fine. but, give it something where there is lots of dependancies and watch it crash and burn.


In the end, what some people are saying here is the truth, if rambus spent half the time they tried attepmting to push their ram via legal methods on more R&D, they might have found that making a superior product is a safer way of trying to push a new product. Simply trying to change things with no benifit for anyone is a futile idea.

I really pity the guys at rambus who was responsible for the companies approach to all this.
 


<< -And there is no need for it in the futre despite claims by some that it will become more disirable when proc speeds increase. Poping in an imaginary 10Ghz proc will not magically make Rambus suddenly shine, in fact it will just make it look worse as the latency penalties will cripple the the enivitable longer pipeline a 10Ghz proc will need. the longer the pipeline gets, the more it can be crippled by high latencies weather it be Rambus or DDR. Which BTW is touching on the major weakness of the P4 in general, the more complexe the task you trough at it, the slower it gets since the whole thing is a high latency model of computing. Sure, give it a simple task like a media file encoding, it will do fine. but, give it something where there is lots of dependancies and watch it crash and burn. >>



You are wrong Xemus. As the pipeline gets longer, it is true that DDR latency will increase, RDRAM on the otherhand, due to their technology, will not. The higher the processor speeds go, RDRAM's performance will actually outperform because of their bandwidth and decreasing latency. Why do you think the advantage of RDRAM over DDR gets bigger as the Mhz goes higher? Look at the increasing benchmark advantages.
 


<< You are wrong Xemus. As the pipeline gets longer, it is true that DDR latency will increase, RDRAM on the otherhand, due to their technology, will not. The higher the processor speeds go, RDRAM's performance will actually outperform because of their bandwidth and decreasing latency. Why do you think the advantage of RDRAM over DDR gets bigger as the Mhz goes higher? Look at the increasing benchmark advantages. >>


The problem is that we are comparing DUAL channel Rdram against single channel DDR. With Intel as well as nvidia coming out and expanding their lines of dual channel pc2700 chipsets, the major performance difference btw the two will be mitigated, with DDR offering lower latency as well as greater bandwidth when compared to the latest Rdram offering.
 



<< You are wrong Xemus. As the pipeline gets longer, it is true that DDR latency will increase, RDRAM on the otherhand, due to their technology, will not. The higher the processor speeds go, RDRAM's performance will actually outperform because of their bandwidth and decreasing latency. Why do you think the advantage of RDRAM over DDR gets bigger as the Mhz goes higher? Look at the increasing benchmark advantages.
>>




memory Latency is a fixxed figure, it does not increase or decrease depending on what the CPU is doing. There seems to be a belief in some circles that rambus will perform better when the pushed harder, I can assure you this is false. I do recall where some of this false piece of information is derivied from, it came from some Intel promotional material which found it's way into one of anand's articles, Anand put this piece of information into one of his articles about 1 year ago and simply added the comment, " look this proves rambus memory will perform better when faster CPU's come along " If anyone really cares, I could find this article in question if asked. The point is, Anand took this material, essentially copied it into his article, and now there is a large group of people who mistakingly take it to be fact. Essentially what this peiece of material in question did was attempt to draw more bandwidth out of SDRAM module than it could do in a certain amount of time, thus resulting in a grossly inflated figure for time for data transfere to complete. This was figure was then used to very falsly make the claim that SDRAM latencies increase when you ask of to much bandwidth from it. What the Intel/RAMBUS PR department conviently left out was that Rambus would do the same thing if you attempted to draw more bandwidth out of it than it was capable of.
 


<<
The problem is that we are comparing DUAL channel Rdram against single channel DDR. With Intel as well as nvidia coming out and expanding their lines of dual channel pc2700 chipsets, the major performance difference btw the two will be mitigated, with DDR offering lower latency as well as greater bandwidth when compared to the latest Rdram offering. >>



Exactly! Although I don't hate RAMBUS, I'm not a big fan either. RDRAM has other issues as well like the "heat" issue (nobody mentioned this in this thread?). I'm not to sure if we'll see RDRAM in P4 notebooks due to the heat generated
 
ssanches: While the heat produced by Rambus is higher than typical memory types, it can hardly be called a 'problem'. nothing a simple process shrink will not fix when they are ready to takle this problem in notebooks.

They don't generate a massive amount of heat as some in the anti-rambus camp will have you believe. As I understand it, most of the chips are actually "off" when not being used. "off" meaning some form of low power / low speed state.


I still think the only real thing against Rambus is still the questionability of it's real performance. We all know Rambus does and can offer more bandwidth, the real question is, is thats what's really needed? My position on this is, no, at least not at the expense of other important performance critaria.
 
The RIMM4200 memory module is way to sweet to hate Rambus anymore. They finally have a product that is impressive to me. I'll let the courts decide the legal issues.
 
If what Sohcan has said about memory latencies is accurate than a 10GHz processor in x86 doesn't make sense with any of today's memory technologies. The speed of the memory would have to ramp with the processor or the memory simply won't cut it. You can only have so many instructions on the fly with x86, right? Well, what happens when you have all of your pipelines stalled from cache misses?

Sure you could interleave the memory to provide more theoretical bandwidth, but you still have timing issues in x86 that make interleaving futile after a certain point. The latency of DDR333 and PC1066 would introduce massive penalties, regardless of interleaving. Caches would have to be near perfect to prevent awful performance penalties.

You would need to basically have an L3 cache running 1GHz+ just to handle stalls when the L1 and L2 caches had a cache miss. You'd have to interleave 4-8 channels of DDR333 or PC1066 memory just to feed the processor in any timely manner for WHEN L3 cache misses occur. Could you imagine the penalty for waiting 3-400 cycles of the processor just to retrieve data from the 1GHz L3 cache, let alone how many cycles to retrieve from your massively-interleaved main memory? Your processor would be starved for data more often than it would be doing work.
 
mad rat, your on the right track...


There are a few ways to look at the solution to this problem.

All solutions boil down to overcoming the increasing dispartity of the CPU speed to the ram speed. xxx_SDRAM or Rmabus will not be able to feed high Ghz procs in the future becuase despite the fact we have lots of tricks with cache, the cpu still sometimes needs data from ram instenatiuosly, if the ram can not do that, the CPU just waits, the higher the Ghz, the more clocks it will waste in doing so. It's not really usefull if the CPU needs one thing now and the Ram says, I will give you 10 things real quick so long as you wait 20 clocks .

Just like a car assembly line, if suddenly a suppliar for parts used on station X is late providing parts, but promises to have sent 3X parts in the next shipment. the Car manufacture is still going to be upset because it will still have a stalled assembly line and in the end prodcues less cars that week. The car manufactures does have a large warehouse of parts ( Cache ), but if for what ever reason the warehouse runs out and the parts suppliar can't provide the goods in a timemly fashion, the end result is no work. And in real life this does happen and people get sent home. The factory is empty , just like your CPU becomes nothing more than a source for High GHZ radio noise

Now say the car maker doubles production and introduces a second assembly line, but again something goes wrong with the supply chain again for what ever reason, now the car maker is going short double the amount of cars.!

now in real life the assembly line usually does not suffer this problem to often, but in a CPU this situation occurs regularly, simply take your favorite CPU cache hit rate, usually in the 95% range, now find the oppisite figure of about 5%, and thats how often your CPU will be in a stalled state on average. Now considering the latency of todays ram, consider how during those nasty cache misses how all those latancies can add up to thousands&thousands of clocks, Of course you will begin to understand thar ram latancy needs t be kept as low as possible, even if that means slowing the over all bandwidth down in order to accomplish this.

Faster CPU clock are simply going to mean more stalls per second. In general we have to accept pipe line stalls, but the real solution will be making them as painless as possible. Rambus does not solve this problem but only makes it worse as CPU speed increases
 
Why don't we hate RAMBUS?.........Uhhhh......Bandwidth maybe. No matter how you slice and dice it, RDRAM800 is still the fastest around. Since prices on DDRAM have risen, price is almost a mute point. Then along comes RDRAM 4500. More bandwidth. Hmmmm...hard to hate that if the price is right. The reality seems to be we're all whores for speed.
 
RAMBUS commited fraud.

They joined the JEDEC and helped to promote industry standards that they knew they held patents on. All companies were required to disclose any pending patents they held on technologies so that the JEDEC would not develop a standard that belonged to any one company. RAMBUS failed to disclose their pending patents, and only left the JEDEC when their patents had cleared the patent office. I fully believe that memory architectures will be driven to a serial format, and that parallel memory will go the way of the dodo. That doesn't mean that I will support a company that prevents competition and chooses to act in a way that is contrary to ethical standards. When RAMBUS cleans up their act, I will buy their products.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top